Prediction: These 2 Artificial Intelligence (AI) Stocks Will Soar After Feb. 26 (Hint: Not Nvidia)

CoreWeave is positioned to beat expectations on its Feb. 26 report driven by a reported $56 billion revenue backlog (up 271% year-over-year at the end of Q3) and analyst estimates that revenue could more than double to about $12 billion in 2026 after a prior $5 billion 2025 guide; Nvidia’s recent $2 billion strategic investment and a partnership to build 5 GW of AI factories by 2030 plus early access to Vera Rubin chips (claimed to cut inference costs up to 10x and training GPU needs 4x) underpin the upside. SoundHound AI, reporting the same day, has seen its share price fall over 65% since October but trades at under 20x sales, with a cumulative subscriptions/bookings backlog of $1.2 billion at end-2024 (up 75% YoY) and traction for its Amelia agentic voice platform that could drive a positive turnaround if backlog and guidance print favorably.

Analysis

Market structure: Nvidia-led acceleration of AI compute commoditizes GPU-powered capacity but concentrates power in GPU suppliers (NVDA) and fast-capacity operators (CRWV). Expect winners: CRWV (capacity reseller scaling backlog: $56B end-Q3 2026) and NVDA (chip vendor), while legacy slow-expanders and small cloud providers face pricing pressure and longer sales cycles; electrical utilities and hyperscale real-estate landlords will see higher demand for power and PPA exposure as data centers target ~5 GW by 2030.

Risk assessment: Key tail risks are export controls or prolonged GPU shortages that could halve projected revenue growth (>50% upside priced into CRWV) and customer-concentration losses (top customers: MSFT/META/OpenAI). Timeline: immediate (Feb 25–26 earnings volatility), short-term (next 3–6 months capacity ramp and Vera Rubin deployments), long-term (2026–2030 structural demand for power and CAPEX). Monitor supply-chain (TSMC/NVDA cadence), backlog conversion rates, and gross margin trends as second-order liquidity drains.

Trade implications: Tactical: favor infra exposure with downside protection — CRWV and NVDA are primary longs; use defined-risk option spreads around earnings to avoid IV crush. Rotate out of lower-growth consumer tech (NFLX) into AI infra and enterprise voice automation (selective SOUN exposure only on confirmed backlog growth >20% y/y). Bonds/FX: strong tech flows may compress IG spreads and support USD; watch power/commodity input inflation for capex cost pressure.

Contrarian angles: Consensus underestimates execution and margin risk for fast-growing infra names — massive backlogs can be supply-constrained, not revenue-guarantees. The market may be underpricing regulatory/export risk which could re-rate multiples by 20–40%. Historical parallel: 2017–18 GPU cycles where supply bottlenecks created volatile earnings; expect similar stop-start revenue recognition and windows to buy dips rather than chase runs.

Key Decisions for Investors

• Establish a 2–3% long position in CRWV ahead of Feb 26 results but hedge with a 3-month collar: buy CRWV shares, purchase 3-month 10% OTM puts (cost-limited) and sell 3-month 25% OTM calls; reason: capture backlog-driven upside while limiting one-off post-earnings gap risk.
• Buy NVDA selectively post-Feb 25 print: target 1–2% portfolio exposure via 3-month call debit spreads (buy ATM call, sell 15% OTM) to limit IV exposure; if NVDA guides materially above consensus (>10% beat in data-center revenue), scale to 3–4% within 48 hours.
• Avoid pre-earnings outright long in SOUN; instead set a conditional buy: allocate 1% position in SOUN only if Q4 report shows backlog growth >20% y/y and forward bookings lead to >20% y/y revenue guidance for 2026; otherwise short-term avoid due to churn and valuation multiple compression risk.
• Implement a sector rotation: reduce consumer/streaming exposure (e.g., trim NFLX by 2–3%) and reallocate into AI infra & enterprise AI software (CRWV, NVDA, selective SOUN) over the next 4–8 weeks; simultaneously add 0.5–1% exposure to utilities/PPAs or energy-transition names benefitting from heavy data-center power demand (time horizon 12–36 months).